Free workflow review
Also called: Discovery call
A free 30-minute call. You bring one workflow that costs time, money or attention, and leave with one candidate workflow, a fit decision and the next step. Nothing is built or run on it.
WhereBook one
26 words from this site in plain English: the phases, the eight checks every build is crashed into, the record you receive and the parts of a workflow.
Also called: Discovery call
A free 30-minute call. You bring one workflow that costs time, money or attention, and leave with one candidate workflow, a fit decision and the next step. Nothing is built or run on it.
WhereBook one
Also called: Discovery, audit
The first paid phase. It fixes the workflow, the outcome, the scope and the stop point before any build money is at risk. You keep the workflow map, baseline, ROI case and priority action plan, even if you stop there.
WhereWhat we build
Linked fromWhat we build
Also called: Implementation
The phase where the automation is built. Every check that applies runs on test data before the system touches real data, and a failing check blocks handover.
WhereWhat we build
Linked fromHow we testWhat we buildProcess
Also called: Hypercare
The 14 days after launch that come with every System Build, before the system is left to run on its own.
WhereWhat we build
Also called: Maintenance plan
An optional monthly phase: monitoring, a monthly health review with the last run date per check, and one improvement cycle. The checks run again after any model or prompt change, and at least every 90 days.
WhereWhat we build
Linked fromWhat we build
Also called: Go or no-go decision
The decision that ends a phase. Say yes and the next phase starts. Say no and it ends there. Every phase has one, and the decision is yours.
WhereFive gates
Linked fromWhat we build
Also called: Sign-off
The written scope is met and every check that applies has passed. Only then is a system handed over.
WhereGate 3
Also called: Transfer
Your team gets the system, the stress-test record, the documentation, the runbook and a named owner. Your team owns the workflow, code, configuration, prompts and data, and production accounts stay under your control.
WhereGate 4
Linked fromHow we testWhat we buildProcess
Also called: Operating guide
A short document that tells your team how to run the system day to day and what to do when a step fails. It is part of every handover, so the system stays operable without us.
WhereWhat we build
Linked fromWhat we buildProcess
Also called: Eval set, golden set, labeled test cases
A fixed set of labeled cases with the right answer written down first. Every result is compared with it, so a change that breaks an answer shows up before it ships.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: Prompt injection
A test input with hidden instructions inside it, such as a lead form that tells the AI to ignore its rules. The workflow must ignore them or flag them.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: Idempotency, replay safety
Sending the same input twice, or rerunning after a crash, never doubles a record, a message or a payment.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: Retries, error handling
A failed call is retried a set number of times. If it still fails, it becomes a parked item for a person. Nothing is dropped silently.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: Load test, rate limits, spend caps
The workflow runs at its own maximum batch size, and it has limits on input size, items per run, sending speed and AI spend.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: Alerting
If a run fails, a named person gets an email saying which step failed.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: Human in the loop, approval step
Nothing reaches a customer, a candidate or a money system without a person, unless a fixed rule allows it.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: Guardrails, output validation
Fixed rules run before the AI step and check its answer after. If the AI fails or is down, the workflow returns its fallback.
WhereThe 8 checks
Linked fromHow we testWhat we buildWhat we automate
Also called: QA record, test report
The record delivered with every System Build: dated, one row per check, counts only, never your data. Each row carries one of the status words.
WhereThe record
Linked fromHow we testWhat we buildProcess
Also called: Test status
The four words a record row can carry. Passed: a dated, recorded run met the check's bar. Not yet: the run has not happened. Not needed: the check does not apply, and the row says why. Stopped: the check stopped the record, and the row says what happened next.
WhereThe record
Also called: Not applicable
A check is marked Not needed for one of four reasons only: there is no AI step, the AI sees only numbers our code computed, the workflow writes and sends nothing, or it stays inside your team.
WhereThe 8 checks
Also called: Model version check
The date each AI step's model was last reviewed. It is recorded for every AI step, and the checks run again before any model swap.
WhereThe AI steps
Linked fromWhat we buildProcess
Also called: Certification
The mark a workflow carries only while its checks are current. It comes off when the last run is older than 90 days, or older than the latest model or prompt change.
WhereHow we test
Also called: LLM step
A step where an AI model reads, judges or drafts. Fixed rules check its answer. When a rule can do the work reliably, there is no AI step.
Also called: Deterministic step, business rule
A step that checks, decides or routes the same way every time, written down in advance. Rules run before and after every AI step.
WhereThe live workflow on home
Linked fromHow we testWhat we automate
Also called: Dead-letter queue, exception queue
An item that still failed after its retries, set aside for a person with the failed step named, so nothing is lost or sent half done.
Also called: Safe default
The safe result a workflow returns when its AI step fails or is down, such as sending the item to a person instead of guessing.
Linked fromHow we testWhat we build
No word here matches that.
Ask about it on the call, or send the question in writing.
Every page below links the first use of a word back to its card here. A filled square is a page where you can read the word in context. An empty row is a word with its own card that no other page links to yet.
Bring one workflow. In 30 minutes you hear which of the eight checks it would face, and whether it is worth building.